Skip to content

Latest commit

 

History

History
135 lines (84 loc) · 4.49 KB

README.md

File metadata and controls

135 lines (84 loc) · 4.49 KB
Raw

Backend

Python service for the InferGPT backend.

Structure

  • /supervisors storing all supervisor modules (agents that call other agents for a greater goal)
  • /agents containing all agents the director can call. Agents have their own functions stored within the agent module
  • /utils with all shared utility modules
  • /tools for all shared function modules

Set up

  1. Ensure the .env file has been configured as described in the main README.

Note: You will need to configure a LLM to run the service.

This README covers instructions on how to run the application:

  • Locally
  • In a Docker Container

For ease of use, we would recommended that you run the entire application using Docker Compose instead. See main README.

If you would prefer not to use Docker Compose, read on...

Unless otherwise stated all of the commands mentioned in this README should be run from ./backend.

Running Locally

Follow the instructions below to run the backend locally. Change directory to /backend, then follow the instructions below.

  1. Set-up a virtual environment

It's recommended (though not technically required) to create a virtual environment for the project by running python -m venv .venv to create it and .venv/Scripts/activate to activate it in your active terminal.

  1. Install dependencies
pip install -r requirements.txt

(VsCode) You may run into some issues with compiling python packages from requirements.txt. To resolve this ensure you have downloaded and installed the "Desktop development with C++" workload from your Visual Studio installer.

  1. Run the app
uvicorn src.api:app --port 8250
  1. Check the backend app is running at http://localhost:8250/health

Running in a Docker Container

  1. Build the Docker image
docker build -t {my-backend-image-name} .
  1. Run the backend within a Docker container
docker run --env-file ../.env -p 8250:8250 {my-backend-image-name}

Note that we pass in the entire environment file that contains all our application's configuration. This means some unneccessary configuration is also being passed in. This is fine when testing locally. In production, we must limit this to only the essential backend configuration (see environment configuration within Docker Compose).

  1. Check the backend app is running at http://localhost:8250/health

Linting

Ruff is being used for the backend linting. See the docs: here

Make sure the ruff dependency is downloaded; It is defined in the requirements.txt so step 1 of setup should do this for you.

Run the following command to check:

ruff --version

If it is not installed use the following command:

pip install ruff

If ruff is installed correctly, the python files in the backend will be checked using the following command, from the backend directory:

ruff check

You can run the following command to fix any errors automatically:

ruff format

The ruff vscode plugin can also be installed from the store to show linting errors in the IDE.

Linting Rules

Currently there are 4 rule groups selected inruff.toml. All rule groups can be found here.

To add further rules, these are added to ruff.toml by using the letter asssigned to the rules as in the docs linked above. ie. pep8-naming uses the letter "N".

Test

pytest is being used for testing the backend. Like with linting, running the setup steps should download pytest for you.

We are using a separate tests directory to store all the tests. This directory is intended to mirror the src directory to make it easier to find the tests for a specific module.

Warning

running the pytest tests only works when running the service locally (not through the Docker setup)

To run the tests, change to the /backend directory and run the following command:

pytest

More documentation on pytest can be found here.

Contributing

Type errors

The vscode extension Pylance can be used to highlight basic type errors in your python code. To set this up do the following:

  • Install the extension
  • Add the following to your ../.vscode/settings.json file:
{
    "python.analysis.typeCheckingMode": "basic",
}